A premature ejaculation control nerve regulation method and system based on multi-mode biosensing and AI algorithm, and a medium

Abstract

This application relates to a neural modulation method, system, and medium for controlling premature ejaculation based on multimodal biosensing and AI algorithms, belonging to the field of communication stimulation control technology. The neural modulation method for controlling premature ejaculation includes: monitoring and collecting and processing different physiological activities of the human perineum to obtain multidimensional physiological response signal clusters; calculating the comprehensive arousal level based on historical multidimensional physiological signals and a preset physical pattern, and constructing an arousal curve model; identifying and judging the multidimensional physiological response signal clusters based on the arousal curve model, determining the type of arousal node, and matching behavioral intervention strategies; updating and optimizing the arousal curve model based on active behavioral actions and implicit feedback mechanisms; and by monitoring the user's biophysiological signals in real time, using AI algorithms to analyze the trend of arousal changes, and performing physical intervention through electrical stimulation at key nodes, thereby reducing sensitivity, delaying the ejaculation threshold, and achieving the goal of prolonging sexual intercourse time.

Description

Technical Field

[0001] This application relates to the field of communication stimulation control technology, and in particular to a neural modulation method, system and medium for controlling premature ejaculation based on multimodal biosensing and AI algorithms. Background Technology

[0002] In recent years, the number of devices used for physical therapy to treat premature ejaculation has gradually increased, mainly including premature ejaculation trainers and electrical pulse stimulation devices. Among them, premature ejaculation trainers are mostly integrated devices or large devices. When using them, the device usually needs to be placed on the penis and kept in an erection. The training cycle is relatively long, and it is often difficult to wear them for a long time in actual use, which limits their practicality. Some large devices also require users to go to the hospital regularly for electrical stimulation treatment, which is time-consuming, frequent, and inconvenient.

[0003] In contrast, electrical pulse stimulation devices release electrical signals through pulse circuits, which act on the sensitive nerves of the dorsal root of the penis, thereby reducing the nerve excitation threshold, inhibiting nerve sensitivity, and effectively prolonging ejaculation time.

[0004] Existing patents disclose a method and device for dynamically adjusting ejaculation control neural modulation via electrical pulses, belonging to the field of medical device electronics. The method includes: Step S1, determining whether the position of the two electrodes used for stimulating the dorsal root nerve of the penis is an effective stimulation position based on the collected contact resistance at the dorsal root of the penis; if yes, proceed to the next step; otherwise, adjust the position of the two electrodes until it is determined to be an effective position before proceeding to the next step; Step S2, selecting at least one working mode from high-frequency mode, low-frequency mode, mixed-frequency mode, mixed-frequency group mode, and modulated mixed-frequency mode, and outputting electrical pulse signals of the corresponding frequency mode to the dorsal root nerve of the penis through the two electrodes for electrical stimulation; Step S3, during the electrical stimulation process, dynamically adjusting the output energy of the electrical pulse signal based on the collected contact resistance value of the two electrodes, controlling the dynamic balance of the charging and discharging process to ensure the effectiveness and safety of the electrical stimulation. The ejaculation control neural modulation of the above invention is safer and more effective.

[0005] The existing technical solutions mentioned above have the following drawbacks: 1. Traditional solutions mostly use fixed modes or intensities, which cannot be dynamically adjusted according to the user's real-time physiological arousal state. This one-size-fits-all stimulation method may fail to effectively prolong the duration due to insufficient intensity, or may affect the user experience due to excessive intensity or inappropriate timing, thus failing to achieve a precise and personalized treatment loop.

[0006] 2. Traditional equipment often relies heavily on manual adjustments and sensory feedback from the user, lacking objective physiological indicators as a basis for intervention. This not only places high demands on the user's operational experience and judgment but also makes the treatment process difficult to standardize, resulting in unstable effects and an inability to continuously optimize through data. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the purpose of this application is to provide a neural modulation method, system, and medium for controlling premature ejaculation based on multimodal biosensing and AI algorithms. By monitoring the user's biophysiological signals in real time, using AI algorithms to analyze the trend of arousal changes, and performing physical intervention through electrical stimulation at key nodes, the method reduces sensitivity, delays the ejaculation threshold, and achieves the goal of prolonging sexual intercourse time.

[0008] This was achieved using the following technical solutions: In a first aspect, this application provides a neural modulation method for controlling premature ejaculation based on multimodal biosensing and AI algorithms, including: Monitoring and collecting data on different physiological activities of the human perineum yields a multidimensional cluster of physiological response signals. Based on historical multidimensional physiological signals and a pre-set physical pattern, the overall arousal level is calculated, and an arousal curve model is constructed. Based on the excitation curve model, identify and judge multidimensional physiological response signal clusters, determine the type of excitation node, and match behavioral intervention strategies; The excitement curve model is updated and optimized based on proactive behaviors and implicit feedback mechanisms.

[0009] By adopting the above technical solution, multidimensional physiological signals of the perineum are collected in real time, and the data is processed using wavelet transform and feature extraction techniques. Combined with historical signals, a personalized arousal curve model is constructed through time series modeling. Furthermore, a pattern recognition algorithm is used to determine the type of arousal node and match corresponding behavioral intervention strategies. Finally, a reinforcement learning mechanism is introduced to continuously optimize the model based on user feedback, thereby achieving accurate monitoring and adaptive intervention of physiological state, significantly improving the personalization level of individual health management and the effectiveness of intervention.

[0010] This application is further configured to: monitor and collect data on different physiological activities of the human perineum to obtain multidimensional physiological response signals, including: The sensor type is used to monitor different physiological activities of the human perineum and collect blood oxygenation signals, body posture inertia signals, skin conductance response signals, surface electromyography signals and skin temperature signals. Based on a preset reference timestamp, the blood oxygenation signal, body posture inertia signal, skin conductance response signal, surface electromyography signal, and skin temperature signal are aligned to construct an initial physiological signal cluster; Based on the signal type, the initial physiological signal cluster is filtered, denoised, and anomaly corrected to obtain a noise-free physiological signal cluster. Time-frequency domain analysis and feature fusion were performed on the noiseless physiological signal clusters to obtain multidimensional physiological response signal clusters.

[0011] By adopting the above technical solution, multi-modal sensors are used to collect multi-dimensional physiological signals such as blood oxygen, body electrophysiology, and electromyography. Signal synchronization and noise reduction are performed through timestamp alignment and adaptive filtering technology. Wavelet transform and feature fusion algorithms are used to extract time-frequency domain features and construct a high-quality multi-dimensional physiological response signal cluster, thereby achieving comprehensive and high-precision monitoring of the physiological state of the perineum.

[0012] This application further specifies: based on historical multidimensional physiological signals combined with a preset physical pattern, calculating the comprehensive arousal level and constructing an arousal curve model, including: Based on a preset feature extraction frequency, historical multidimensional physiological signals are extracted in a sliding manner, and multidimensional physiological response feature values ​​are calculated. The multidimensional physiological response feature values ​​are normalized based on the feature baseline values ​​to obtain the multidimensional standard feature values. Based on the preset physical constitution pattern m, the multidimensional standard feature values ​​are classified and aggregated, and the pattern similarity weight is calculated. Weighting of pattern similarity based on time series The overall excitability is calculated by weighting and fusing the multidimensional standard eigenvalues. ; Where t is the timestamp, i is the physiological characteristic number, and N is the total number of physiological characteristics. The standard feature value of the i-th physiological characteristic in the n-th time window; Smoothing regression and sequence transformation are performed on the overall excitement level to generate an excitement curve; The excitability curves of the same physical pattern are segmented according to the window time limit to obtain the window feature sequence; Quantitative fitting analysis is performed on the window feature sequence to extract feature variation nodes and calculate the time series feature regression degree; Peak and trough detection is performed on the excitation curve, and a feature node variation fitting curve is constructed by combining the feature variation nodes. The threshold for determining the type of excitatory node is determined by combining the fitted curve of feature node changes with the time-series feature regression degree. An excitability curve model is constructed by iteratively training the excitability node determination threshold based on historical behavior action sequences.

[0013] By adopting the above technical solution, historical multidimensional physiological signal features are extracted and normalized through sliding window, similarity weights are calculated by combining preset physical patterns, and a comprehensive excitability curve is generated using a time-series weighted fusion algorithm. Then, an excitability model with adaptive thresholds is constructed through smooth regression, peak and valley detection, and feature node fitting analysis. Finally, the model parameters are iteratively optimized based on historical behavioral data, which realizes accurate quantification and dynamic prediction of individual physiological excitability state, significantly improving the personalization and real-time performance of state recognition.

[0014] This application is further configured to: identify and determine multidimensional physiological response signal clusters based on an excitability curve model, determine the type of excitability node, and match behavioral intervention strategies, including: Based on the analytical judgment layer of the excitation curve model, the multidimensional physiological response signal cluster is decomposed to obtain the signal node amplitude and calculate the node change rate. If the amplitude of the signal node is greater than the preset resting threshold, and the node change rate is positive within the preset time window, then the current excitability node is determined to be the excitation initiation node. If the rate of change of a node is the local maximum within the current time window, then the current excitability node is determined to be an excitability maxima node. If the node change rate is the global maximum value in the current cycle, but less than the preset ejaculation threshold, then the current excitability node is determined to be the node with the highest global excitability. If the node change rate is negative, but the absolute value is a local maximum within the current time window, then the current excitability node is determined to be an excitability minimum node. If the amplitude of the signal node is within the tolerance range of the resting threshold, then the current excitatory node is determined to be an excitatory-to-calm node. Based on the strategy matching layer of the excitation curve model, and combined with the type of excitation node, the corresponding behavioral intervention strategy is matched. If it is the initial point of excitation, then a progressive single intervention stimulus is performed according to the intervention priority, and the intensity of the stimulus is increased as the intervention time increases; If it is the point of maximum excitation, then switch the intervention level and implement a strong double stimulus intervention to interrupt the accumulation of excitation. If it is the highest point of global arousal, then full-dimensional stimulation suppression is implemented, and the ejaculation reflex is inhibited by combining external cues with multi-dimensional stimulation combinations. If it is a node of minimum excitation, then staged stimulation intervention is performed. As the amplitude of the signal node decreases, the intervention level is gradually switched and the stimulation combination is modified until the current node is reached, and then stimulation is stopped. If the node is in a state of excitement returning to calm, then standby monitoring is performed, and a low-power mode is initiated.

[0015] By adopting the above technical solution, based on real-time signal processing and dynamic threshold determination algorithms, the amplitude and rate of change of physiological signal nodes are analyzed in real time through the excitation curve model. Five types of excitation nodes, such as initial rise and maximum value, are accurately identified. Differentiated behavioral intervention strategies are adaptively matched according to the node type, thereby achieving millisecond-level response and personalized control of physiological excitation state, significantly improving the accuracy, timeliness and user experience of intervention.

[0016] This application further specifies: updating and optimizing the excitement curve model based on active behaviors combined with implicit feedback mechanisms, including: Monitor users' proactive behaviors and actions, record proactive adjustment operations, and mark the exact timestamps of their occurrence; Based on the exact timestamp of occurrence and the degree of adjustment, the active adjustment operation is quantified, and the true arousal level is calculated. Based on the excitement curve model combined with the exact timestamp of occurrence Calculate and predict excitement levels; Calculate the active performance error value based on the actual and predicted excitement levels. ; Based on the active operation error value combined with the excitation curve model, the excitation feature vector The weight vector matrix of the excitation curve model The optimized weight matrix is ​​obtained by performing optimization and correction. ; in, The learning rate; A global dynamic learning rate mechanism is constructed by adjusting the learning rate during the update and optimization process based on implicit feedback. Based on the global dynamic learning rate mechanism, the modified weight matrix is ​​back-derived and interpolated to optimize the virtual true value excitation curve. The virtual true excitation curve and the predicted excitation curve are solved globally based on the overall deviation threshold, and the optimal weight vector matrix is ​​calculated.

[0017] By adopting the above technical solution, based on the implicit feedback adaptive learning algorithm, the true arousal level is calculated by quantifying the user's active adjustment behavior and comparing it with the model's predicted value to obtain the error. The gradient descent method is used to optimize the model weight matrix. At the same time, the dynamic learning rate mechanism and virtual truth curve construction are combined to optimize the global parameters, thereby realizing online personalized calibration and continuous optimization of the arousal prediction model, significantly improving the model's adaptability to individual physiological response patterns and long-term prediction accuracy.

[0018] This application further specifies: a global dynamic learning rate mechanism is constructed by adjusting the learning rate during the update and optimization process based on an implicit feedback mechanism, including: Based on the type of active behavior and the error value of active operation, the implicit feedback events in the implicit feedback mechanism are classified into levels, the level of feedback events is determined, and a basic learning rate coefficient is assigned. Statistical analysis is performed on the active operation error value to quantify the error fluctuation value, and the instantaneous fluctuation measure is calculated by combining the signal-to-noise ratio of physiological characteristic signals. The sign of the active operation error value is determined, and the continuity consistency factor is calculated; Based on the continuity consistency factor Instantaneous fluctuation measurement and base learning rate coefficient Initial learning rate during the update and optimization process Feedback adjustment is performed to obtain the dynamic learning rate. ; ; in, This is the gain coefficient; A global dynamic learning rate mechanism is constructed by setting boundaries and introducing decay factors for the dynamic learning rate.

[0019] By adopting the above technical solution, based on the adaptive optimization algorithm, feedback events are classified and assigned a basic learning rate by analyzing user behavior type and operation error. Then, the learning rate is dynamically adjusted by integrating multi-dimensional information such as error fluctuation statistics, signal-to-noise ratio and error sign consistency. Furthermore, a global dynamic learning rate strategy is constructed by combining boundary constraints and decay mechanisms. This enables intelligent adaptive adjustment of the learning rate during model update, effectively improving convergence stability and personalized adaptability.

[0020] Secondly, this application also provides a neuromodulation system for controlling premature ejaculation based on multimodal biosensing and AI algorithms, employing the following technical solution: A neural modulation system for controlling premature ejaculation based on multimodal biosensing and AI algorithms, comprising the following methods for implementing neural modulation for controlling premature ejaculation: The data acquisition and execution terminal is used to monitor and process different physiological activities of the human perineum, obtain multidimensional physiological response signal clusters, and identify and judge the type of excitatory nodes based on the excitability curve model to match behavioral intervention strategies. The control intervention end is used to update and optimize the excitement curve model based on proactive behaviors and implicit feedback mechanisms, and to correct behavioral intervention strategies. The processing interface is used to calculate the overall arousal level and construct an arousal curve model based on historical multidimensional physiological signals combined with preset physical patterns.

[0021] By adopting the above technical solution, based on multi-sensor fusion and adaptive learning algorithms, the physiological signals of the perineum are collected and processed in real time and features are extracted. Personalized arousal curves are constructed using time series modeling. Then, the intervention strategy is dynamically matched with pattern recognition. At the same time, an implicit feedback mechanism is introduced to continuously optimize the model parameters, realizing a closed-loop system from physiological monitoring and state recognition to intelligent intervention, which significantly improves the accuracy, real-time performance and personalized adaptability of the intervention.

[0022] Thirdly, this application also provides an electronic device, comprising: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by one or more processors, the one or more processors implement any of the methods in the above scheme.

[0023] Fourthly, this application also provides a storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the premature ejaculation control neural modulation method based on multimodal biosensing and AI algorithm as described above.

[0024] In summary, the beneficial technical effects of this application are as follows: By using multi-sensor fusion and adaptive learning algorithms, physiological signals of the perineum are collected and processed in real time and features are extracted. Personalized arousal curves are constructed using time series modeling, and then intervention strategies are dynamically matched by pattern recognition. At the same time, an implicit feedback mechanism is introduced to continuously optimize model parameters, realizing a closed-loop system from physiological monitoring and state recognition to intelligent intervention, which significantly improves the accuracy, real-time performance and personalized adaptability of the intervention. By analyzing user behavior types and operational errors, feedback events are classified and assigned a base learning rate. Then, multi-dimensional information such as error fluctuation statistics, signal-to-noise ratio, and error sign consistency are integrated to dynamically adjust the learning rate. Furthermore, a global dynamic learning rate strategy is constructed by combining boundary constraints and decay mechanisms. This enables intelligent adaptive adjustment of the learning rate during model updates, effectively improving convergence stability and personalized adaptability. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the neuromodulation method for controlling premature ejaculation in this application; Figure 2 This is a flowchart illustrating step S46 of the premature ejaculation control neuromodulation method in this application; Figure 3 This is a schematic diagram of the neural regulation system for controlling premature ejaculation in this application. Detailed Implementation

[0026] The present application will be further described in detail below with reference to the accompanying drawings.

[0027] Reference Figure 1 This application discloses a neural modulation method for controlling premature ejaculation based on multimodal biosensing and AI algorithms, comprising: S1: Monitor and collect data on different physiological activities of the human perineum to obtain a multidimensional physiological response signal cluster; S2: Based on historical multidimensional physiological signals and a preset physical pattern, calculate the overall arousal level and construct an arousal curve model; S3: Identify and judge multidimensional physiological response signal clusters based on the excitation curve model, determine the type of excitation node, and match behavioral intervention strategies; S4: Update and optimize the excitement curve model based on active behaviors and implicit feedback mechanisms.

[0028] In this embodiment, after the user wears a smart wearable device that integrates multimodal sensors such as blood oxygen, skin conductance, and electromyography, the device first collects real-time physiological signals from the perineal area through multiple sensors simultaneously. After filtering, alignment, and feature fusion processing, a multidimensional physiological response signal cluster is formed. At the same time, based on the user's historical training data and personal physical condition, the device uses a sliding window feature extraction and time-series weighted fusion algorithm to calculate a personalized comprehensive arousal level and constructs a unique arousal curve model for the user.

[0029] During training, the model analyzes current physiological signals in real time, identifies key node types such as the initial rise of excitement and the maximum value of excitement, and automatically matches corresponding behavioral intervention strategies. For example, when the initial rise of excitement node is detected, the device will start progressive stimulation and gradually increase the intensity. If the highest global excitement node close to the ejaculation threshold is identified, it will immediately switch to full-dimensional suppressive stimulation to delay the orgasm reflex.

[0030] Users can also actively adjust the device according to their own feelings through the buttons. These operations are recorded and the actual excitement level is calculated. By comparing with the model's predicted value, the excitement curve model is optimized online based on the implicit feedback mechanism and dynamic learning rate algorithm, so that the next node identification and intervention strategy is more in line with the user's physiological response characteristics, thereby achieving sustainable personal adaptive rehabilitation training.

[0031] Preferably, step S1 includes: The sensor type is used to monitor different physiological activities of the human perineum and collect blood oxygenation signals, body posture inertia signals, skin conductance response signals, surface electromyography signals and skin temperature signals. Based on a preset reference timestamp, the blood oxygenation signal, body posture inertia signal, skin conductance response signal, surface electromyography signal, and skin temperature signal are aligned to construct an initial physiological signal cluster; Based on the signal type, the initial physiological signal cluster is filtered, denoised, and anomaly corrected to obtain a noise-free physiological signal cluster. Time-frequency domain analysis and feature fusion were performed on the noiseless physiological signal clusters to obtain multidimensional physiological response signal clusters.

[0032] In this embodiment, N-dimensional physiological time-series signals are synchronously collected through wearable devices. Common dimensions include: ECG / heart rate variability: reflecting autonomic nervous activity; skin conductance response: reflecting sympathetic nerve excitability; body temperature: reflecting metabolic state; triaxial acceleration: reflecting the amount of physical activity and used for noise reduction; and electroencephalography (EEG, if applicable): directly reflecting the cortical excitation state.

[0033] Signal purification methods include bandpass filtering (to remove power frequency interference and baseline drift) and wavelet transform (to remove motion artifacts); outliers are eliminated or corrected by interpolation within a physiologically reasonable range (e.g., heart rate between 40-200 bpm). All signal streams are unified to the same timestamp and sampling frequency.

[0034] Multidimensional biosensors include: PPG sensor: monitors heart rate (HR) and blood oxygen saturation (Sp2); IMU motion sensor: monitors the frequency, amplitude, and positional changes of pelvic movements; EDA / GSR (electrodermal response sensor): monitors skin conductivity, which is the most sensitive indicator reflecting sympathetic nerve excitation (emotional / sexual arousal); EMG (electromyography sensor): monitors the tension and contraction frequency of the pelvic floor muscles (PC muscles), which is directly related to the urgency of ejaculation; and Temp (high-precision body temperature sensor): monitors changes in local skin temperature, reflecting the state of congestion.

[0035] Preferably, step S2 includes: Based on a preset feature extraction frequency, historical multidimensional physiological signals are extracted in a sliding manner, and multidimensional physiological response feature values ​​are calculated. The multidimensional physiological response feature values ​​are normalized based on the feature baseline values ​​to obtain the multidimensional standard feature values. Based on the preset physical constitution pattern m, the multidimensional standard feature values ​​are classified and aggregated, and the pattern similarity weight is calculated. Weighting of pattern similarity based on time series The overall excitability is calculated by weighting and fusing the multidimensional standard eigenvalues. ; Where t is the timestamp, i is the physiological characteristic number, and N is the total number of physiological characteristics. The standard feature value of the i-th physiological characteristic in the n-th time window; Smoothing regression and sequence transformation are performed on the overall excitement level to generate an excitement curve; The excitability curves of the same physical pattern are segmented according to the window time limit to obtain the window feature sequence; Quantitative fitting analysis is performed on the window feature sequence to extract feature variation nodes and calculate the time series feature regression degree; Peak and trough detection is performed on the excitation curve, and a feature node variation fitting curve is constructed by combining the feature variation nodes. The threshold for determining the type of excitatory node is determined by combining the fitted curve of feature node changes with the time-series feature regression degree. An excitability curve model is constructed by iteratively training the excitability node determination threshold based on historical behavior action sequences.

[0036] In this embodiment, time-domain / frequency-domain / nonlinear feature extraction is used: features are calculated for each signal window (e.g., every 5 minutes).

[0037] Heart rate variability: RMSSD (parasympathetic activity) and LF / HF ratio (sympathetic / parasympathetic balance) were extracted; Skin conductance: SCR frequency and amplitude were extracted; Body temperature: Change relative to baseline was extracted; Activity level: Root mean square value was extracted.

[0038] All feature values ​​are Z-scre standardized based on the individual’s historical baseline (such as the mean and standard deviation of the past 7 days at rest) to eliminate absolute differences between individuals and scaled to the [0,1] interval; Based on medical or empirical models, several constitution patterns are defined (such as "sympathetic-dominant," "balanced," and "parasympathetic-sensitive"). Each pattern is characterized by: different sensitivities to different physiological signals; different characteristic baselines at rest; and by analyzing an individual's historical signal response patterns under resting and standard stimuli, they are classified into the closest predefined constitution pattern, or their similarity weights with each pattern are calculated using algorithms.

[0039] Slender build (Sensitive): Characteristics: Soft waveform, high frequency, narrow pulse width. Suitable for: Users with high nerve sensitivity and low tolerance.

[0040] Standard: Features: Balanced waveform combination, medium intensity and frequency. Suitable for: Most general users.

[0041] Endurance: Features: Deep stimulation waveform, low-frequency wide pulse, high intensity limit. Suitable for: Users with well-developed muscles, high tolerance, or severe desensitization needs.

[0042] The current overall excitement index is calculated using a weighted fusion algorithm: E(t)=w_1*Nrm(GSR_t)+w_2*Nrm(HR_t)+w_3*Nrm(EMG_t)+w_4*Activity(ACC_t); HR_t: Real-time heart rate; GSR_t: Skin conductivity (core indicator of excitability); EMG_t: Pelvic floor muscle tone; ACC_t: Pelvic motion acceleration; Fusion feature E(t) (comprehensive excitability index).

[0043] Smoothing the {E(t)} sequence by applying Savitzky-Golay filtering or LOWESS local regression yields a smooth excitation curve.

[0044] Optionally, a hidden Markov model or variational autoencoder can be used to model the curve and learn its typical variation patterns.

[0045] Identify local peaks (excitement peaks) and troughs (relaxation troughs) in the curve. Divide the curve into different state intervals based on preset thresholds or clustering algorithms, such as: High excitation period: E(t)>θ_high; Calm period: θ_low≤E(t)≤θ_high; Low excitation / fatigue period: E(t)<θ_low.

[0046] Preferably, step S3 includes: Based on the analytical judgment layer of the excitation curve model, the multidimensional physiological response signal cluster is decomposed to obtain the signal node amplitude and calculate the node change rate. If the amplitude of the signal node is greater than the preset resting threshold, and the node change rate is positive within the preset time window, then the current excitability node is determined to be the excitation initiation node. If the rate of change of a node is the local maximum within the current time window, then the current excitability node is determined to be an excitability maxima node. If the node change rate is the global maximum value in the current cycle, but less than the preset ejaculation threshold, then the current excitability node is determined to be the node with the highest global excitability. If the node change rate is negative, but the absolute value is a local maximum within the current time window, then the current excitability node is determined to be an excitability minimum node. If the amplitude of the signal node is within the tolerance range of the resting threshold, then the current excitatory node is determined to be an excitatory-to-calm node. Based on the strategy matching layer of the excitation curve model, and combined with the type of excitation node, the corresponding behavioral intervention strategy is matched. If it is the initial point of excitation, then a progressive single intervention stimulus is performed according to the intervention priority, and the intensity of the stimulus is increased as the intervention time increases; If it is the point of maximum excitation, then switch the intervention level and implement a strong double stimulus intervention to interrupt the accumulation of excitation. If it is the highest point of global arousal, then full-dimensional stimulation suppression is implemented, and the ejaculation reflex is inhibited by combining external cues with multi-dimensional stimulation combinations. If it is a node of minimum excitation, then staged stimulation intervention is performed. As the amplitude of the signal node decreases, the intervention level is gradually switched and the stimulation combination is modified until the current node is reached, and then stimulation is stopped. If the node is in a state of excitement returning to calm, then standby monitoring is performed, and a low-power mode is initiated.

[0047] In this embodiment, if the comprehensive excitation index E(t) exceeds the resting threshold and shows a continuous upward trend, it is determined to be the start of excitability increase (T_start), and a gradual intervention is performed: electrical stimulation is initiated, and the current intensity is gradually increased to allow the nervous system to adapt.

[0048] If the rate of change of the comprehensive excitation index E'(t) reaches a local maximum (i.e., the excitation increases the fastest), it is determined to be the node with the fastest excitation increase (T_max_slope_up), and a dual strong intervention is performed: electrical stimulation and vibration are applied simultaneously to interrupt the accumulation of excitation and achieve an emergency "brake".

[0049] If the comprehensive arousal index E(t) reaches a local peak and is close to the ejaculation threshold, it is determined to be the highest point of arousal (T_peak), and full-dimensional suppression is implemented: first, a ring vibration is used as a cue, and then a triple combination of electrical stimulation, vibration and infrared heat therapy is used to suppress the ejaculation reflex.

[0050] If the overall excitability index E(t) begins to decline, it is determined to be a T-descent process, and a phased intervention is implemented: stimulation is maintained in the early stage of decline to prevent rebound; when the rate of change E'(t) reaches the maximum downward slope (i.e. the fastest decline), electrical stimulation is stopped immediately.

[0051] The starting point of excitability increase (T_{start}): Definition: E(t) exceeds the resting threshold and E'(t)>0 for a specific time window.

[0052] Strategy: Initiate mild electrical stimulation (preheating) to prompt the nervous system to adapt.

[0053] The node with the maximum excitability-enhancing slope angle (T_{max_slope_up}): Definition: When E'(t) reaches a local maximum, it represents a sharp increase in excitation.

[0054] Strategy: Strong intervention. Increase the frequency and intensity of stimulation to interrupt the rapid accumulation of excitation signals and apply the "brake".

[0055] Peak of excitation (T_{peak}): Definition: E(t) reaches a local peak in the current cycle, close to the ejaculation threshold.

[0056] Strategy: Maintain or pulse interference. Use high-frequency short pulse interference to prevent exceeding the threshold, while the ring vibrates to prompt the user to adjust the amplitude of their movements.

[0057] The node with the maximum slope down angle of excitability descent (T_{max_slope_down}): Definition: When E'(t) is negative and has the largest absolute value, it indicates that the intervention is effective and the excitement level drops rapidly.

[0058] Strategy: Reduce stimulation. Gradually decrease the intensity of electrical stimulation to avoid excessive inhibition that could lead to loss of erection.

[0059] Excitatory-to-calm node (T_{rest}): Definition: E(t) regresses to near the baseline.

[0060] Strategy: Standby / weak maintenance. Enter monitoring mode and prepare for the next cycle.

[0061] Preferably, step S4 includes: S41: Monitor users' proactive behaviors and actions, record proactive adjustment operations, and mark the exact timestamp of the occurrence; S42: Based on the exact timestamp of occurrence and the degree of adjustment, quantify the active adjustment operation and calculate the true excitability; S43: Based on the excitement curve model combined with the exact timestamp of occurrence Calculate and predict excitement levels; S44: Calculate the active operation error value based on the actual excitement level and the predicted excitement level. ; S45: Based on the active operation error value combined with the excitation curve model, the excitation feature vector The weight vector matrix of the excitation curve model The optimized weight matrix is ​​obtained by performing optimization and correction. ; in, The learning rate; S46: Based on the implicit feedback mechanism, the learning rate is adjusted during the update and optimization process to construct a global dynamic learning rate mechanism; S47: Based on the global dynamic learning rate mechanism, the modified weight matrix is ​​reverse-engineered and interpolated to construct a virtual true value excitement curve; S48: Based on the overall deviation threshold, perform a global solution for the virtual true excitation curve and the predicted excitation curve, and calculate the optimal weight vector matrix.

[0062] In this embodiment, while real-time monitoring of physiological signals Xt=[GSR,HR,EMG,ACC] and calculation of the comprehensive arousal index E(t), all active adjustment operations of the user (such as manually increasing / decreasing stimulation intensity, performing emergency stop, etc.) and their exact timestamps t are continuously listened to and recorded. action .

[0063] The "implicit feedback" to the current state of true excitement, and quantified into a "truth value" excitement level E. true (t action ).

[0064] For example, "manually increasing intensity" implies that the user perceives the current stimulus as insufficient, potentially underestimating arousal levels, therefore E true (t actionThe value should be higher than the currently calculated E(t). action ).

[0065] Calculate the prediction error δ of the model at that moment: δ=E true (t action )−E(t action This error directly reflects the deviation between the model's predictions and the user's actual experience.

[0066] For each weight component w i (Adjustments will be made for GSR, HR, EMG, and ACC respectively): When the user manually increases the intensity (underestimation, δ>0), for physiological features with higher current values ​​(larger xi(t)), the corresponding weight w i This will result in a positive increase, making the model more sensitive to similar high-value features in the future.

[0067] When the user manually reduces the intensity (overestimation, δ<0), the weights of features with higher current values ​​will be reduced to correct the model's tendency to overestimate.

[0068] If a "failed ejaculation interception" event occurs, it is classified as the most serious false negative. This triggers a significant weight update, typically by increasing the learning rate η or directly applying a large correction to the weights, to quickly improve the model's sensitivity to warning signals.

[0069] At the same time, lowering the excitation threshold of key intervention points (such as T_start) will enable interventions to be triggered earlier in the future, thus avoiding missed detections again.

[0070] Based on all user-initiated actions and their timestamps recorded during this period, a continuous "virtual truth excitability curve" covering the entire treatment course is constructed through reverse deduction and interpolation.

[0071] With the goal of minimizing the overall deviation between the predicted curve and the virtual true curve (e.g., using the least squares method), a new set of optimal weight configurations is calculated.

[0072] The weights obtained from online learning are combined with the optimal weights calculated offline (such as by weighted averaging) and smoothly updated into the main model as the initial weights for the next treatment cycle, thereby achieving the gradual evolution of the model.

[0073] Missed detection rate: The proportion of cases where intervention should have been initiated but was not (e.g., near ejaculation but without strong intervention). If this rate increases, the threshold for early interventions such as the onset of heightened arousal (T_start) will be automatically lowered to make it more sensitive.

[0074] False trigger rate: The proportion of unnecessary interventions (such as a user being given a stronger stimulus while at rest). If this rate increases, the threshold for aggressive interventions such as double strong interventions (T_max_slope_up) will be automatically raised to make them more conservative.

[0075] Implicit feedback mechanisms include: Actively increasing intensity → the system underestimates excitability → high-value feature weights need to be increased.

[0076] Actively reducing intensity → The system overestimates excitement level → It is necessary to reduce the weight of high-value features.

[0077] Uninterrupted ejaculation → serious missed detection → significantly increased weight, reduced trigger threshold.

[0078] Preferably, refer to Figure 2 Step S46 includes: A: Based on the type of active behavior and the error value of active operation, the implicit feedback events in the implicit feedback mechanism are classified into levels, the level of feedback events is determined, and a basic learning rate coefficient is assigned. B: Perform statistical analysis on the active operation error value, quantify the error fluctuation value, and combine it with the signal-to-noise ratio of physiological characteristic signals to calculate the instantaneous fluctuation measure; C: Determine the sign of the active operation error value and calculate the continuity consistency factor; D: Based on the continuity consistency factor Instantaneous fluctuation measurement and base learning rate coefficient Initial learning rate during the update and optimization process Feedback adjustment is performed to obtain the dynamic learning rate. ; ; in, This is the gain coefficient; E: By setting boundaries and introducing decay factors for the dynamic learning rate, a global dynamic learning rate mechanism is constructed.

[0079] In this embodiment, implicit feedback events are classified according to the type of user's proactive behavior and the severity of the implicit error, and a base learning rate multiplier (β) is assigned to each level: Level S (Severe Missed Event): Such as "ejaculation not intercepted". This event indicates that the model has completely failed and needs to be learned as quickly as possible. Set the base multiplier β_S (e.g., 5.0).

[0080] Level A (Strong Correction Event): Such as "User executes emergency stop (not manual adjustment)". This indicates that the model deviates significantly from the true state and requires rapid correction. Set the base multiplier β_A (e.g., 2.5).

[0081] Level B (Explicit Adjustment Event): Such as "manually increasing / decreasing intensity significantly." This indicates a clear bias in the model, requiring significant adjustment. Set the base multiplier β_B (e.g., 1.5).

[0082] Level C (Fine-tuning event): such as "Manual slight adjustment of intensity". This indicates that the model is close but not precisely matched, requiring fine-tuning. Set the base multiplier β_C (e.g., 0.8).

[0083] Level D (No event or confirmed event): such as "user did not act" or "system intervention was successful and user was not interrupted". This indicates that the model prediction may be reliable and should enter the decay or fine-tuning phase. Set the base multiplier β_D (e.g., 0.2).

[0084] We introduce a metric U(t) that reflects the model's "confidence" at the current moment, used to inversely adjust the learning rate (the more uncertain the model, the higher the learning rate should be). This metric can be based on: Volatility of recent prediction errors: For example, calculating the standard deviation of the absolute value of the error |δ| over the past N seconds.

[0085] Signal-to-noise ratio or quality indicators of characteristic signals: For example, whether a channel signal (such as ACC) shows drastic fluctuations may indicate noise interference.

[0086] The above factors are normalized to form a value U(t)∈[0,1], where 1 represents extreme uncertainty.

[0087] To prevent model oscillations caused by a single abnormal operation or user accidental touch, a continuous consistency factor C(t) is introduced.

[0088] If a user's recent consecutive actions all convey the same corrective direction (e.g., three consecutive manual increases in intensity), it indicates that the model bias is consistent, and reinforcement learning should be implemented. C(t) can be calculated based on the sign consistency of the error δ over the past M valid operations. For example, if all signs are the same, then C(t) = 1.5; if there are frequent alternations between positive and negative signs, then C(t) = 0.5.

[0089] Each time implicit feedback is received and a weight update is required, the dynamic learning rate η_dynamic(t) used for this update is calculated.

[0090] To ensure numerical stability, a safety constraint is imposed on η_dynamic: Set upper and lower bounds: η_min ≤ η_dynamic ≤ η_max. For example, η_min can be set to 0.01, and η_max can be set to 10 times the initial η_base.

[0091] Introducing a decay factor: During periods of calm when no user feedback is received for an extended period (dominated by D-level events), the system can periodically apply a slight decay (e.g., multiply by 0.99) to η_dynamic, causing it to slowly regress to the base value η_base.

[0092] Reference Figure 3 A neural modulation system for controlling premature ejaculation based on multimodal biosensing and AI algorithms, applied to a neural modulation method for controlling premature ejaculation, includes: The data acquisition and execution terminal is used to monitor and process different physiological activities of the human perineum, obtain multidimensional physiological response signal clusters, and identify and judge the type of excitatory nodes based on the excitability curve model to match behavioral intervention strategies. The control intervention end is used to update and optimize the excitement curve model based on proactive behaviors and implicit feedback mechanisms, and to correct behavioral intervention strategies. The processing interface is used to calculate the overall arousal level and construct an arousal curve model based on historical multidimensional physiological signals combined with preset physical patterns.

[0093] In this embodiment, the acquisition and execution end is a smart electrical stimulation patch worn in the perineum, which is the core sensing and execution unit of the system; it acts on the perineal nerve branches through flexible electrodes and outputs a microcurrent with a specific waveform to interfere with the transmission of excitation signals.

[0094] The control intervention device is a smart ring controller worn on the finger, providing discreet and convenient real-time control. The intensity of electrical stimulation can be adjusted via a knob or side touch strip. It can handle strong interference or emergency cessation in cases of sudden hyperexcitability. Micro-vibrations provide users with notifications of the current device status or AI intervention prompts.

[0095] Users can manually adjust and save the following parameters according to their own feelings: frequency (Hz), pulse width (µs), waveform type (sine wave / square wave / triangle wave / combined wave), and stimulation rhythm (continuous / intermittent / burst).

[0096] After the user wears a portable monitoring device that integrates sensors for skin conductance, heart rate variability, and surface electromyography, the system first acquires multidimensional physiological signals of the perineal and pelvic floor muscles under stress in real time through the acquisition execution end. After filtering, alignment, and feature fusion, a signal cluster reflecting the excitability of the autonomic nervous system is formed.

[0097] Meanwhile, the processing terminal generates a personalized stress excitability curve model based on the user's historical physiological data over the past three months and the "high-response" physical pattern determined through questionnaire assessment. This model is used to extract features from a sliding window and a weighted fusion algorithm. In actual high-pressure tasks (such as emergency surgery or flight simulation), the system dynamically identifies the node type of physiological signals through this model. For example, when a rapid increase in the amplitude of the skin conduction signal is detected and the rate of change is positive, it is determined to be a "stress activation node," which then triggers the intervention strategy matched by the acquisition and execution terminal. This could be done by outputting low-frequency mindfulness guidance pulses through bone conduction headphones or by applying regular low-frequency vibrations to the perineal area through a bionic tactile feedback device to promote accessory nerve activation.

[0098] If a user feels that the intervention intensity is insufficient or excessive, they can actively adjust it via the wristband button (such as increasing vibration or changing the guided language). At this time, the control intervention terminal will record the timestamp and adjustment range of the operation, and dynamically compare the deviation between the actual physiological response and the model prediction using an implicit feedback mechanism. The weight parameters of the excitation curve model are corrected online using an adaptive learning rate algorithm, and the subsequent node recognition threshold and intervention strategy matching logic are optimized accordingly. After several weeks of closed-loop training, the system gradually adapts to the user's specific physiological response pattern, and finally realizes the ability to automatically provide personalized and pre-adaptive neuromodulation intervention in real high-pressure environments, effectively improving the user's psychological stability and rapid recovery ability under extreme load.

[0099] An electronic device, comprising: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by one or more processors, the one or more processors implement any of the methods in the above scheme.

[0100] A storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the premature ejaculation control neural modulation method as described above.

[0101] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A neural modulation method for controlling premature ejaculation based on multimodal biosensing and AI algorithms, characterized in that, include: Monitoring and collecting data on different physiological activities of the human perineum yields a multidimensional cluster of physiological response signals. Based on historical multidimensional physiological signals and a pre-defined physical pattern, the overall arousal level is calculated, and an arousal curve model is constructed. Based on the excitation curve model, the multidimensional physiological response signal clusters are identified and judged to determine the excitation node type and match behavioral intervention strategies. The excitement curve model is updated and optimized based on proactive behaviors and implicit feedback mechanisms.

2. The neural modulation method for controlling premature ejaculation based on multimodal biosensing and AI algorithms according to claim 1, characterized in that, The monitoring and data acquisition of different physiological activities in the human perineum yields multidimensional physiological response signals, including: The sensor type is used to monitor different physiological activities of the human perineum and collect blood oxygenation signals, body posture inertia signals, skin conductance response signals, surface electromyography signals and skin temperature signals. The blood oxygenation signal, the body posture inertia signal, the skin conductance response signal, the surface electromyography signal, and the skin temperature signal are aligned according to a preset reference timestamp to construct an initial physiological signal cluster; The initial physiological signal cluster is filtered, denoised, and anomaly corrected according to the signal type to obtain a noise-free physiological signal cluster. Time-frequency domain analysis and feature fusion were performed on the noiseless physiological signal cluster to obtain a multidimensional physiological response signal cluster.

3. The neural modulation method for controlling premature ejaculation based on multimodal biosensing and AI algorithms according to claim 1, characterized in that, The process of calculating comprehensive arousal level and constructing an arousal curve model based on historical multidimensional physiological signals and a preset physical pattern includes: Based on a preset feature extraction frequency, historical multidimensional physiological signals are extracted in a sliding manner, and multidimensional physiological response feature values ​​are calculated. The multidimensional physiological response feature values ​​are normalized based on the feature baseline values ​​to obtain multidimensional standard feature values. The multidimensional standard feature values ​​are categorized and aggregated according to the preset physical constitution pattern, and the pattern similarity weight is calculated. The pattern similarity weights and the multidimensional standard feature values ​​are weighted and fused according to the time series to calculate the comprehensive excitement level.

4. The neural modulation method for controlling premature ejaculation based on multimodal biosensing and AI algorithms according to claim 3, characterized in that, The method of calculating comprehensive arousal level and constructing an arousal curve model based on historical multidimensional physiological signals and a preset physical pattern also includes: The overall excitement level is subjected to smooth regression and sequence transformation to generate an excitement level curve; The excitability curves of the same physical pattern are segmented according to the window time limit to obtain the window feature sequence; The window feature sequence is subjected to quantitative fitting analysis to extract feature variation nodes and calculate the time series feature regression degree; Peak and trough detection is performed on the excitation curve, and a feature node change fitting curve is constructed by combining the feature variation nodes. Based on the fitted curve of the feature node change and the time-series feature regression degree, the threshold for determining the excitability node type is determined. The excitability node determination threshold is iteratively trained based on the historical behavior sequence to construct an excitability curve model.

5. The neural modulation method for controlling premature ejaculation based on multimodal biosensing and AI algorithms according to claim 1, characterized in that, The step of identifying and judging the multidimensional physiological response signal clusters based on the excitation curve model, determining the type of excitation node, and matching behavioral intervention strategies includes: Based on the analytical judgment layer of the excitation curve model, the multidimensional physiological response signal cluster is decomposed to obtain the signal node amplitude and calculate the node change rate. If the amplitude of the signal node is greater than the preset resting threshold, and the node change rate is positive within the preset time window, then the current excitability node is determined to be the excitation initiation node. If the rate of change of the node is the local maximum within the current time window, then the current excitability node is determined to be an excitability maxima node. If the node change rate is the global maximum value in the current cycle, but less than the preset ejaculation threshold, then the current excitability node is determined to be the node with the highest global excitability. If the node's rate of change is negative, but its absolute value is a local maximum within the current time window, then the current excitability node is determined to be an excitability minimum node. If the amplitude of the signal node is within the tolerance range of the resting threshold, then the current excitatory node is determined to be an excitatory-to-calm node.

6. The neural modulation method for controlling premature ejaculation based on multimodal biosensing and AI algorithms according to claim 1 or 5, characterized in that, The step of identifying and judging the multidimensional physiological response signal clusters based on the excitation curve model, determining the type of excitation node, and matching behavioral intervention strategies further includes: Based on the strategy matching layer of the excitation curve model, and combined with the type of excitation node, the corresponding behavioral intervention strategy is matched. If it is the initial point of excitation, then a progressive single intervention stimulus is performed according to the intervention priority, and the intensity of the stimulus is increased as the intervention time increases; If it is the point of maximum excitation, then switch the intervention level and implement a strong double stimulus intervention to interrupt the accumulation of excitation. If it is the highest point of global arousal, then full-dimensional stimulation suppression is implemented, and the ejaculation reflex is inhibited by combining external cues with multi-dimensional stimulation combinations. If it is a node of minimum excitation, then a phased stimulation intervention is performed. As the amplitude of the signal node decreases, the intervention level is gradually switched, and the stimulation combination is modified until the current node is reached, and then the stimulation is stopped. If the node is in a state of excitement returning to calm, then standby monitoring is performed, and a low-power mode is initiated.

7. The neural modulation method for controlling premature ejaculation based on multimodal biosensing and AI algorithms according to claim 1, characterized in that, The process of updating and optimizing the excitement curve model based on active behaviors and implicit feedback mechanisms includes: Monitor users' proactive behaviors and actions, record proactive adjustment operations, and mark the exact timestamps of their occurrence; Based on the exact timestamp of occurrence and the degree of adjustment, the active adjustment operation is quantified to calculate the true degree of excitement. The predicted excitement level is calculated based on the excitement curve model combined with the exact timestamp of occurrence; Calculate the active operation error value based on the actual excitement level and the predicted excitement level; Based on the active operation error value and the excitation feature vector of the excitation curve model, the weight vector matrix of the excitation curve model is optimized and corrected to obtain the corrected weight matrix. A global dynamic learning rate mechanism is constructed by adjusting the learning rate during the update and optimization process based on implicit feedback. Based on the global dynamic learning rate mechanism, the modified weight matrix is ​​reverse-engineered and interpolated to construct a virtual true value excitation curve. The virtual true excitation curve and the predicted excitation curve are globally solved based on the overall deviation threshold to calculate the optimal weight vector matrix.

8. The neural modulation method for controlling premature ejaculation based on multimodal biosensing and AI algorithms according to claim 7, characterized in that, The step of adjusting the learning rate during the update and optimization process based on implicit feedback to construct a global dynamic learning rate mechanism includes: Based on the type of active behavior and the error value of active operation, the implicit feedback events in the implicit feedback mechanism are classified into levels, the level of feedback events is determined, and a basic learning rate coefficient is assigned. The active operation error value is statistically analyzed to quantify the error fluctuation value, and the instantaneous fluctuation metric is calculated by combining the signal-to-noise ratio of physiological characteristic signals. The sign of the active operation error value is determined, and the continuity consistency factor is calculated; The initial learning rate during the update and optimization process is adjusted based on the continuous consistency factor, the instantaneous fluctuation metric, and the basic learning rate coefficient to obtain the dynamic learning rate. A global dynamic learning rate mechanism is constructed by setting boundaries and introducing decay factors for the dynamic learning rate.

9. A neural modulation system for controlling premature ejaculation based on multimodal biosensing and AI algorithms, used to implement the neural modulation method for controlling premature ejaculation as described in any one of claims 1-8, characterized in that, include: The data acquisition and execution terminal is used to monitor and process different physiological activities of the human perineum, obtain multidimensional physiological response signal clusters, and identify and judge the type of excitatory nodes based on the excitability curve model to match behavioral intervention strategies. The control intervention end is used to update and optimize the excitement curve model based on active behavior and implicit feedback mechanisms, and to correct the behavioral intervention strategy. The processing interface is used to calculate the overall arousal level and construct an arousal curve model based on historical multidimensional physiological signals combined with preset physical patterns.

10. A storage medium storing at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the premature ejaculation control neuromodulation method as described in any one of claims 1 to 8.

Images